2

I'm working on an analysis using a negative binomial mixed effects model using glmmTMB, and using R Markdown to generate a report for my collaborators. When I knit to html, the regression output changes slightly. Also, AIC, BIC, logLik and deviance all are shown as NA, although they are not NA before I knit.

I tried setting a seed, but it made no difference. When I knit to pdf, results do not change upon knitting - the problem seems to be specific to knitting to html. Any thoughts?

Andrea Hildebrand
  • 21
  • 1
  • 2
    That's extremely surprising. Is there any chance you can post a reproducible example? There would be an outside chance that threading + a highly optimized compiler could change floating-point outcomes in a non-reproducible way (see [here](https://www.mail-archive.com/r-help@r-project.org/msg267756.html)), but that would only happen if (1) you were using more than one core *and* (2) the model was very unstable (at least to cause a change to a non-positive-definite Hessian that gives NA values for AIC etc.) – Ben Bolker Aug 21 '23 at 01:20

0 Answers0